Data Vizualisation Portfolio
Hugo Marthinet
2024-12-16
This is my data visualization portfolio.
I gathered code I wrote to create some nice plots, put them in an Rmarkdown to create an HTML page where I would have an easy and centralized access to all of those code lines. The goal is to preview the graphs so I can identify the features I want to re-use while creating a new plot, and to be able to copy and paste the chunks I want.
The .Rmd file can run by itself, all the data used to draw the plots are either simulated, arbitrarily defined, or from a dataset.
Below is the list of all the required packages. Mostly graphical tools, along with some statistical analysis resources.
# Tidy data
library(tidyverse)
library(lubridate)
# Supplementary analysis
library(survival)
library(survminer)
library(TraMineR)
library(rstatix)
library(epiR)
# Data
library(ISLR)
# plotly
library(plotly)
# ggplot2 and addins
library(ggplot2)
library(ggfortify)
library(ggpubr)
library(gridExtra)
library(ggseqplot)
library(ggh4x)
library(ggstatsplot)
# Other graphics tools
library(treemap)1 Basics
Starting with the basics : histograms, barplots, pie charts…
Basic is good, but make it pretty.
1.1 Barplots
1.1.1 Frequencies
An horizontal and stacked barplot :
# Simulate data for a barplot
dataBarplot <- data.frame(
Episode = rep(c("First Covid-19 episode", "Last Covid-19 episode"), each = 63), # Two Covid episodes
Time = factor(rep(rep(c("Acute phase", "3 months or more", "Still persisting"), each = 21), 2),
levels = c("Acute phase", "3 months or more", "Still persisting")), # Symptoms last for a given time
Symptoms = rep(c("Abdominal pain", "Anxiety", "Chest pain", "Cognitive disfunction", "Cough", # List of possible symptoms
"Depression", "Dizziness", "Fatigue", "Fever", "Gastrointestinal problem",
"Headache", "Joint pain", "Loss of smell or taste", "Menstruation issue", "Neuralgia",
"New allergies", "Shortness of breath", "Sleeping disorder", "Tachycardia", "Tinnitus",
"Troubled vision"), 6),
N = c(0, 0, 0, 0, 2, 5, 9, 285, 404, 714, 646, 870, 765, 1873, 2632, 2106, 3181, 3844, # Number of cases
3985, 4921, 3529, 0, 0, 0, 0, 0, 0, 3, 38, 51, 49, 82, 116, 119, 287, 142, 448,
190, 336, 158, 81, 502, 0, 1, 4, 6, 12, 12, 6, 315, 243, 192, 232, 212, 930, 394,
394, 902, 781, 351, 460, 92, 1318, 0, 0, 0, 0, 0, 0, 0, 71, 62, 119, 148, 162,
217, 123, 427, 347, 520, 669, 825, 730, 574, 0, 0, 0, 0, 1, 0, 0, 3, 7, 13, 7, 10,
22, 8, 13, 40, 15, 34, 18, 17, 36, 0, 1, 1, 2, 1, 2, 2, 61, 67, 59, 51, 58, 85,
217, 84, 200, 172, 79, 18, 115, 296)) %>%
# Total number of cases per symptoms and episode
mutate(Ntot = c(rep(by(N[Episode == "First Covid-19 episode"], Symptoms[Episode == "First Covid-19 episode"], sum), 3),
rep(by(N[Episode == "Last Covid-19 episode"], Symptoms[Episode == "Last Covid-19 episode"], sum), 3)))
# Initialize the plot with the dataset (DATA_ex2_1)
ggplot(dataBarplot) +
# Add a stacked bar chart (geom_col creates bars with heights determined by 'N')
geom_col(
mapping = aes(N, reorder(Symptoms, -N), fill = Time), # Map 'N' to x, 'Symptome' to y, and 'Temps' to fill color
width = 0.6, # Set the bar width
position = "stack" # Stack bars by the 'fill' variable
) +
# Create facets (separate panels) for each level of the 'Episode' variable
facet_grid(cols = vars(Episode)) +
# Add a vertical line at x = 0 for reference
geom_vline(aes(xintercept = 0)) +
# Customize the x-axis scale
scale_x_continuous(
limits = c(0, 6500), # Set the range for the x-axis
breaks = seq(0, 6000, 1000), # Major tick marks every 1000 units
minor_breaks = seq(500, 6500, 500), # Minor tick marks every 500 units
expand = c(0, 0) # Remove padding on the x-axis
) +
scale_fill_manual(
name = "Symptoms duration",
values = c('#28AFB0', '#F6803C', '#82354F')
) +
# Add text labels to display 'Ntot' values next to each bar
geom_text(
mapping = aes(Ntot + 25, y = Symptoms, label = Ntot), # Position and content of the text
hjust = 0, # Align text to the left of its position
nudge_x = 1 # Slightly nudge text horizontally
) +
# Add titles and subtitles (empty here for now)
labs(title = "", subtitle = "") +
# Customize the overall appearance of the plot
theme(
axis.title = element_blank(), # Remove axis titles
panel.background = element_rect(fill = "white"), # Set panel background to white
panel.grid.major.x = element_line(color = "darkgray"), # Major grid lines for x-axis
panel.grid.minor.x = element_line(colour = "lightgray"), # Minor grid lines for x-axis
legend.position = "bottom"
)
This one is vertical and dodged :
# Simulate data for a barplot
dataBarplot <- data.frame(
Time = factor(rep(c("Acute phase", "3 months or more", "Still persisting"), each = 14),
levels = c("Acute phase", "3 months or more", "Still persisting")), # Symptoms last for a given time
Symptoms = rep(c("Fatigue", "Loss of smell or taste", "Menstruation issue", "Neuralgia", "Fever", "Gastrointestinal problem",
"Headache", "Sleeping disorder", "Tachycardia", "Joint pain",
"New allergies", "Shortness of breath", "Tinnitus",
"Troubled vision"), 3),
N = c(285, 404, 714, 646, 870, 765, 1873, 2632, 2106, 3181, 3844, # Number of cases
3985, 4921, 3529, 38, 51, 49, 82, 116, 119, 287, 142, 448,
190, 336, 158, 81, 502, 315, 243, 192, 232, 212, 930, 394,
394, 902, 781, 351, 460, 92, 1318))
# Initialize the plot with the dataset (DATA_ex2_1)
ggplot(dataBarplot) +
# Add a stacked bar chart (geom_col creates bars with heights determined by 'N')
geom_col(
mapping = aes(reorder(Symptoms, -N), N, fill = Time), # Map 'N' to x, 'Symptome' to y, and 'Temps' to fill color
width = 0.9, # Set the bar width
position = "dodge" # Stack bars by the 'fill' variable
) +
# Add a vertical line at x = 0 for reference
geom_hline(aes(yintercept = 0)) +
# Customize the x-axis scale
scale_y_continuous(
limits = c(0, 5500), # Set the range for the x-axis
breaks = seq(0, 5000, 1000), # Major tick marks every 1000 units
minor_breaks = seq(500, 5500, 500), # Minor tick marks every 500 units
expand = c(0, 0) # Remove padding on the x-axis
) +
scale_fill_manual(
name = "Symptoms duration",
values = c('#28AFB0', '#F6803C', '#82354F')
) +
# Add text labels to display 'Ntot' values next to each bar
geom_text(
mapping = aes(x = Symptoms, y = N, label = N), # Position and content of the text
nudge_x = rep(c(-0.3, 0, 0.3), each = 14),
nudge_y = 120,
size = 2
) +
# Add titles and subtitles (empty here for now)
labs(title = "", subtitle = "") +
# Customize the overall appearance of the plot
theme(
axis.title = element_blank(), # Remove axis titles
panel.background = element_rect(fill = "white"), # Set panel background to white
panel.grid.major.y = element_line(color = "darkgray"), # Major grid lines for x-axis
panel.grid.minor.y = element_line(colour = "lightgray"), # Minor grid lines for x-axis
legend.position = "bottom",
axis.text.x = element_text(
size = 10, # Font size
angle = 45, # Rotate text 45 degrees
vjust = 1, # Vertical alignment
hjust = 1 # Horizontal alignment
)
)
1.1.2 Proportions
I like to display percentages as stacked barplots, to make it obvious that values sum up to a hundred.
This one has and grid according to the status, and I feel like it is easier to read the proportions differences whenever the bars are horizontal :
# Simulate data for a bar plot
dataBarplot <- data.frame(
# Create a "Status" column with repeated categories (12 repetitions for each category)
Status = rep(c("Family member", "Active worker", "Retired worker"), each = 12),
# Create an "Affection" column, categorizing Covid-19 status, with ordered factor levels
Affection = factor(
rep(rep(c("No Covid-19", "Covid-19", "Long Covid-19"), each = 4), 3),
levels = c("No Covid-19", "Covid-19", "Long Covid-19")
),
# Create a "Social_satisfaction" column, representing satisfaction levels, with ordered factor levels
Social_satisfaction = factor(
rep(c("Very satisfied", "Satisfied", "Unsatisfied", "Very unsatisfied"), 9),
levels = c("Very satisfied", "Satisfied", "Unsatisfied", "Very unsatisfied")
),
# Define a numeric vector "N" representing counts corresponding to the combinations above
N = c(65, 255, 70, 16, 38, 131, 41, 7, 21, 123, 59, 10,
819, 2532, 591, 130, 640, 1723, 357, 38, 235, 1134, 468, 68,
1444, 4212, 677, 128, 528, 1198, 164, 20, 160, 575, 166, 27)
) %>%
# Add new columns using dplyr's mutate function
mutate(
# Compute the total count (Ntot) for each "Affection" group within each "Status" group
Ntot = c(
# For "Family member", calculate totals per "Affection" group
rep(by(N[Status == "Family member"], Affection[Status == "Family member"], sum), each = 4),
# For "Active worker", calculate totals per "Affection" group
rep(by(N[Status == "Active worker"], Affection[Status == "Active worker"], sum), each = 4),
# For "Retired worker", calculate totals per "Affection" group
rep(by(N[Status == "Retired worker"], Affection[Status == "Retired worker"], sum), each = 4)
),
# Calculate percentages (P) for each count (N) relative to the total count (Ntot) per group
P = N / Ntot * 100
)
# Initialize the plot with the dataset (DATA_ex2_2)
ggplot(dataBarplot) +
# Add a stacked bar chart
geom_col(
mapping = aes(P, Affection, fill = Social_satisfaction), # Map 'P' to x, 'COVID_long_3mod' to y, and 'Q_E1' to fill
width = 0.6, # Set the bar width
position = "stack", # Stack bars by the 'fill' variable
color = "white" # Add white border to bars
) +
# Create facets (separate rows) for each level of the Status variable
facet_grid(rows = vars(Status)) +
# Add vertical lines at x = 0 and x = 100 for reference
geom_vline(aes(xintercept = 0)) +
geom_vline(aes(xintercept = 100)) +
# Customize the x-axis scale
scale_x_continuous(
limits = c(0, 100.2), # Set the range for the x-axis
breaks = seq(0, 100, 10), # Major tick marks every 10%
minor_breaks = seq(5, 95, 10), # Minor tick marks every 5%
expand = c(0, 0), # Remove padding on the x-axis
labels = paste0(seq(0, 100, 10), "%") # Append "%" to the tick labels
) +
# Add titles and subtitles (empty here for now)
labs(title = " ", subtitle = "") +
# Customize the overall appearance of the plot
theme(
axis.title = element_blank(), # Remove axis titles
# axis.text.x = element_blank(), # Uncomment to hide x-axis labels
# axis.text.y = element_blank(), # Uncomment to hide y-axis labels
panel.background = element_rect(fill = "white"), # Set panel background to white
panel.grid.major.x = element_line(color = "darkgray"), # Major grid lines for x-axis
panel.grid.minor.x = element_line(colour = "lightgray") # Minor grid lines for x-axis
) +
# Customize the fill legend for the bar chart
scale_fill_manual(
"Social interactions", # Title for the legend
values = rev(RColorBrewer::brewer.pal(4, "RdYlGn")), # Reverse color palette
labels = c( # Define labels for legend items
"Very satisfied", # Very satisfactory
"Satisfied", # Somewhat satisfactory
"Unsatisfied", # Somewhat unsatisfactory
"Very unsatisfied" # Very unsatisfactory
)
)
Adding some percentages labels and handling long variables names :
# Create the data frame for the bar plot
dataBarplot <- data.frame(
# "LTI" column with long-term illness categories repeated for each age group
LTI = factor(rep(
c("Leukaemia", "Diabetes (insipidus & mellitus)", "Epilepsy", "Multiple Sclerosis",
"Parkinson & affiliated", "Mental Illness"), 5),
levels = c("Leukaemia", "Diabetes (insipidus & mellitus)", "Epilepsy", "Multiple Sclerosis",
"Parkinson & affiliated", "Mental Illness")
),
# "Age" column with age group categories repeated for each illness type
Age = factor(rep(
c("Under 18", "18 to 24 years old", "25 to 44 years old", "45 to 59 years old", "Over 60"), each = 6),
levels = c("Under 18", "18 to 24 years old", "25 to 44 years old", "45 to 59 years old", "Over 60")
),
# Randomly generated values for the "N" column representing counts
N = sample(100:300, 30)
) %>%
# Calculate the percentage contribution of each count (N) within each illness type (LTI)
mutate(
percent = as.numeric(unlist(by(N, LTI, function(x) {x / sum(x) * 100})))[rep(seq(1, 26, 5), 5) + rep(0:4, each = 6)],
# Label for percentages, shown only if the value exceeds 2%
lab = ifelse(percent > 2, paste0(formatC(percent, 1, format = "f"), "%"), "")
)
# Initialize a vector for positioning text labels within each bar
x_text <- rep(NA, nrow(dataBarplot))
# Calculate the cumulative y-position for placing the text labels
for (i in unique(dataBarplot$LTI)) {
for (j in 1:sum(dataBarplot$LTI == i)) {
ifelse(j == 1,
x_text[which(dataBarplot$LTI == i)[j]] <- dataBarplot$percent[dataBarplot$LTI == i][j] / 2,
x_text[which(dataBarplot$LTI == i)[j]] <- sum(dataBarplot$percent[dataBarplot$LTI == i][1:(j-1)]) +
dataBarplot$percent[dataBarplot$LTI == i][j] / 2)
}
}
dataBarplot$x_text <- 100 - x_text # Adjust text position to align with the plot
# Generate the bar plot
ggplot(data = dataBarplot) +
# Create stacked bars for proportions by age group
geom_col(
mapping = aes(x = LTI, y = percent, fill = Age),
width = 0.6, # Bar width
position = "stack", # Stacked bar position
color = "white" # Bar border color
) +
# Add percentage labels to each segment of the bar
geom_text(
aes(x = LTI, y = x_text, label = lab),
size = 2, # Font size
colour = "white" # Text color
) +
# Add titles and axis labels
labs(
title = "Long term illnesses per age group", # Title
ylab = "Proportion", # Y-axis label
xlab = "" # X-axis label (empty)
) +
# Customize the y-axis scale
scale_y_continuous(
breaks = seq(0, 100, 25), # Tick intervals
labels = c("0%", "25%", "50%", "75%", "100%") # Custom tick labels
) +
# Customize the fill colors for the "Age" categories
scale_fill_manual(
values = c('#1E5471', '#28AFB0', '#F6803C', '#82354F', '#E0D6FF'),
name = "Classe ATC1" # Legend title
) +
# Apply a minimal theme to the plot
theme_minimal() +
# Customize the legend and x-axis text
theme(
legend.position = "right", # Position the legend on the right
axis.text.x = element_text(
size = 10, # Font size
angle = 45, # Rotate text 45 degrees
vjust = 1, # Vertical alignment
hjust = 1 # Horizontal alignment
)
)
1.1.3 Incidence rates
A very basic one. But adding a second axis with error-bars representing the incidence rate CI95% :
# Define maximum y-axis value and coefficient for secondary y-axis scaling
ymax <- 200
coeff <- 70 / 200
# Create data frame for bar plot
dataBarplot <- data.frame(
year = 2012:2022, # Years from 2012 to 2022
n_cases = c(120, 87, 78, 96, 117, 153, 139, 164, 181, 188, 163), # Number of cases per year
pop = rep(300000, 11) # Constant population size of 300,000
) %>%
mutate(
# Calculate incidence rate (IR) per 100,000 population
IR = n_cases / pop * 100000,
# Calculate confidence intervals for the incidence rate
IR_lower = epi.conf(cbind(n_cases, pop), ctype = "inc.rate", method = "exact")[, 2] * 100000,
IR_upper = epi.conf(cbind(n_cases, pop), ctype = "inc.rate", method = "exact")[, 3] * 100000
)
# Create the plot
ggplot(dataBarplot, aes(x = year, y = n_cases, width = 0.95)) +
# Add bars for number of cases
geom_col(
mapping = aes(fill = "Incidence"),
fill = "#AAC0AF" # Bar color
) +
# Add a line for incidence rate (IR)
geom_line(
aes(
x = year,
y = IR,
colour = "Incidence rate" # Legend label for line
),
linetype = 1, # Solid line
size = 1, # Line thickness
colour = "black" # Line color
) +
# Add error bars for confidence intervals of the incidence rate
geom_errorbar(
aes(
x = year,
ymin = IR_lower, # Lower bound of the confidence interval
ymax = IR_upper, # Upper bound of the confidence interval
width = 0.2 # Error bar width
)
) +
# Add text labels for the number of cases above each bar
geom_text(
aes(x = year, y = n_cases, label = n_cases),
nudge_y = 7, # Offset text above the bars
size = 6 # Font size
) +
# Customize the y-axis
scale_y_continuous(
expand = c(0, 0), # Remove padding around the axis
limits = c(0, ymax), # Set y-axis limits
breaks = seq(0, ymax, 25), # Major tick marks
name = "Incidence", # Left y-axis label
sec.axis = sec_axis(
trans = ~ . * coeff, # Transformation for secondary axis
name = "Incidence rate", # Right y-axis label
breaks = seq(0, ymax * coeff, 10) # Tick marks for secondary axis
)
) +
# Customize the x-axis
scale_x_continuous(
breaks = 2012:2022, # Tick marks for each year
labels = 2012:2022, # Labels for each year
name = "" # Empty label for the x-axis
) +
# Apply a classic theme
theme_classic() +
# Customize grid lines, axis text, and axis colors
theme(
panel.grid.major.y = element_line(colour = "grey"), # Major grid lines
panel.grid.minor.y = element_line(colour = "lightgray"), # Minor grid lines
axis.text.x = element_text(size = 11), # Font size for x-axis labels
axis.line.y.left = element_line(color = "#AAC0AF"), # Color for left y-axis line
axis.title.y.left = element_text(color = "#AAC0AF"), # Color for left y-axis title
axis.text.y.left = element_text(color = "#AAC0AF") # Color for left y-axis labels
) +
# Add titles and remove legend title
labs(
title = "", # Plot title
fill = "", # Legend title (empty)
x = " ", # x-axis title (empty)
y = " " # y-axis title (empty)
)
1.2 Line charts
Let’s start with a pretty basic double axis line chart :
# Create a dataset for line chart visualization
dataLinechart <- data.frame(
year = seq.Date(as.Date("2015-01-01"), as.Date("2023-01-01"), by = "year"), # Sequence of years from 2015 to 2023
n_tlc = c(545268, 668497, 551234, 619147, 582365, 854256, 968214, 1023578, 945127), # Number of teleconsultations per year
prop_disp = c(0.23, 0.25, 0.24, 0.28, 0.25, 0.35, 0.38, 0.41, 0.37) # Proportion of teleconsultations with dispensing
) %>%
mutate(
n_tlc_disp = n_tlc * prop_disp # Calculate teleconsultations with dispensing
)
# Define the maximum value for y-axis scaling
ymax <- 1250000
# Create the plot using ggplot2
ggplot(dataLinechart, aes(x = year)) +
# Add line and points for teleconsultations with dispensing
geom_line(aes(y = n_tlc_disp, color = "Teleconsultations with dispensing", lty = "Teleconsultations with dispensing")) +
geom_point(aes(x = year, y = n_tlc_disp, color = "Teleconsultations with dispensing"), size = 1.0) +
# Add line and points for total teleconsultations
geom_line(aes(y = n_tlc, color = "Teleconsultations", lty = "Teleconsultations")) +
geom_point(aes(x = year, y = n_tlc, color = "Teleconsultations"), size = 1.0) +
# Add line and points for the ratio (scaled proportion)
geom_line(aes(y = (prop_disp * ymax), color = "Ratio", lty = "Ratio")) +
geom_point(aes(x = year, y = (prop_disp * ymax), color = "Ratio"), size = 1.0) +
# Set labels for the legend
labs(color = "", lty = "") +
# Customize the colors for each line
scale_colour_manual(values = c(
"Teleconsultations with dispensing" = "#be123c",
"Teleconsultations" = "#1e3a8a",
"Ratio" = "#64748b"
)) +
# Customize the line types for each line
scale_linetype_manual(values = c(
"Teleconsultations with dispensing" = 1,
"Teleconsultations" = 1,
"Ratio" = 2
)) +
# Adjust the legend to remove unnecessary symbols
guides(linetype = guide_legend(override.aes = list(shape = NA))) +
# Set up the y-axis with dual scales (primary and secondary)
scale_y_continuous(
breaks = seq(0, ymax, 250000), # Primary axis breaks
labels = formatC(seq(0, ymax, 250000), digits = 0, format = "f", big.mark = " "), # Format for primary axis
sec.axis = sec_axis(
~ . / ymax, # Scale the secondary axis relative to ymax
name = "Ratio of the number of teleconsultations \nwith dispensing to the total number of dispensations",
labels = c("0%", "25%", "50%", "75%", "100%") # Labels for the secondary axis
),
expand = c(0, 0)
) +
# Set the range for the y-axis
coord_cartesian(ylim = c(0, ymax)) +
# Customize the x-axis with year labels
scale_x_date(
breaks = seq.Date(as.Date("2015-01-01"), as.Date("2023-01-01"), by = "year"),
labels = format.Date(seq.Date(as.Date("2015-01-01"), as.Date("2023-01-01"), by = "year"), "%Y")
) +
# Add labels for axes
labs(
x = "",
y = "Number of teleconsultations"
) +
# Apply a minimal theme and customize the plot appearance
theme_minimal() +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
axis.line.x = element_line(colour = "black", size = 0.3),
axis.ticks.x = element_line(colour = "black", size = 0.3),
axis.line.y = element_line(colour = "black", size = 0.3),
axis.ticks.y = element_line(colour = "black", size = 0.3),
legend.position = c(0.25, 0.85), # Position the legend
legend.background = element_rect(fill = "transparent", color = "white") # Add transparent background for legend
)
1.3 Boxplot
I don’t really enjoy boxplots as I feel they get quite boring. But you can add some twists to make it interesting !
1.3.1 Classic
Basic setup :
# Create a data frame for boxplot visualization
dataBoxplot <- data.frame(
med_id = rep(1:8, c(8, 3, 3, 5, 7, 6, 4, 11)), # Assign medical operator IDs, repeated based on surgery counts
op_id = c(1:8, 1:3, 1:3, 1:5, 1:7, 1:6, 1:4, 1:11), # Assign operation IDs for each surgery
setup = pmax(10 - rpois(47, lambda = 4), 1) # Simulate 'setup' scores, capped at 10 and minimum of 1
)
# Create a data frame for the line plot (average setup values per operation ID)
dataLine <- data.frame(
op_id = 1:max(dataBoxplot$op_id), # Sequence of operation IDs
setup = as.numeric(by(dataBoxplot$setup, dataBoxplot$op_id, mean)) # Compute mean 'setup' scores by op_id
)
# Generate the plot using ggplot2
ggplot(dataBoxplot) +
# Add a boxplot layer for setup scores grouped by operation ID
geom_boxplot(
aes(group = op_id, x = op_id, y = setup) # Map operation ID and setup scores to boxplot
) +
# Add a red line connecting the mean setup values for each operation ID
geom_line(
data = dataLine, # Use the data frame with mean setup values
aes(op_id, setup), # Map operation ID to x and mean setup to y
color = "red" # Set line color to red
) +
# Add red points for the mean setup values
geom_point(
data = dataLine, # Use the data frame with mean setup values
aes(op_id, setup), # Map operation ID to x and mean setup to y
color = "red", # Set point color to red
shape = 16 # Use solid circle for points
) +
annotate(
"text",
x = 1:max(dataBoxplot$op_id), y = 10,
label = paste0("(n=", table(dataBoxplot$op_id),")")
) +
# Add a subtitle with the p-value from a linear model (setup ~ op_id)
labs(subtitle = paste0("p=", round(summary(lm(setup ~ op_id, dataBoxplot))$coefficient[2, 4], 4))) +
# Label the y-axis
ylab("Surgery setup grade (out of 10)") +
# Label the x-axis
xlab("Number of surgeries") +
# Customize the y-axis scale
scale_y_continuous(
limits = c(0, 10.5), # Set y-axis limits between 0 and 10
breaks = 0:10, # Add breaks at each integer value
expand = c(0, 0) # Remove padding on the axis
) +
# Customize the x-axis scale
scale_x_discrete(
limits = 1:max(dataBoxplot$op_id), # Limit x-axis to the range of operation IDs
breaks = 1:max(dataBoxplot$op_id) # Add breaks at each operation ID
) +
# Apply a minimal theme for clean visualization
theme_minimal()
Jitter boxplots fixes one of the classic boxplots downside by adding the points to give a better idea of the data distribution :
dataBoxplot <- data.frame(
y = rnorm(200),
Group = sample(LETTERS[1:3], size = 200, replace = TRUE)
)
# Box plot by group with jitter
ggplot(dataBoxplot, aes(x = Group, y = y, colour = Group)) +
geom_boxplot(outlier.shape = NA) + # Box plot without showing outliers
geom_jitter(width = 0.2) + # Add jittered points for individual observations
theme_minimal() + # Use a minimal theme for clean visualization
xlab("Group") + # X-axis label
ylab("Some randomly generated values") # Y-axis label
Adding some two-by-two testing :
# Simulate data for a boxplot
dataBoxplot <- data.frame(
id = rep(1:47, 5), # 47 subjects, repeated 5 times for each group
times = rep(paste0("Month ", c(0, 3, 6, 9, 12)), each = 47), # 5 time points
val = c( # Simulate values for each time point with different means and SDs
rnorm(n = 47, mean = 2, sd = 1), # Group 1 (mean=2, sd=1)
rnorm(n = 47, mean = 5, sd = 1), # Group 2 (mean=5, sd=1)
rnorm(n = 47, mean = 5, sd = 2), # Group 3 (mean=5, sd=2)
rnorm(n = 47, mean = 8, sd = 2), # Group 4 (mean=8, sd=2)
rnorm(n = 47, mean = 8, sd = 5) # Group 5 (mean=8, sd=5)
)
)
dataBoxplot$val <- pmax(0, dataBoxplot$val)
# Perform ANOVA to test differences between the groups
res.aov <- anova_test(
data = dataBoxplot,
dv = val, # Dependent variable: val (size)
wid = id, # Within-subjects factor: id
between = times # Between-subjects factor: times
)
# Pairwise t-tests to compare each group with group 1, adjusting for multiple comparisons using Bonferroni correction
pwc <- dataBoxplot %>%
pairwise_t_test(
val ~ times, # Dependent variable 'val' across levels of 'times'
paired = TRUE, # Paired samples
ref.group = "Month 0", # Use "n°1" as the reference group
p.adjust.method = "bonferroni" # Apply Bonferroni correction to p-values
) %>%
add_xy_position(x = "times") # Add xy-position for displaying p-values
# Create the boxplot with individual data points and p-values
ggboxplot(
dataBoxplot, # Data for the boxplot
x = "times", # Group by 'times'
y = "val", # Plot 'val' (size) on the y-axis
add = "point", # Add individual data points to the boxplot
ylab = "Size", # Label for the y-axis
xlab = "" # No label for the x-axis
) +
scale_y_continuous(
limits = c(0, 25),
expand = c(0, 0)
) +
stat_pvalue_manual(pwc) + # Add p-values from pairwise t-tests
labs(
subtitle = get_test_label(res.aov, detailed = FALSE), # Subtitle for ANOVA result
caption = "2 by 2 testing: Student's t-test, p-value adjustment: Bonferroni" # Caption explaining test details
)
1.3.2 Violins
Switching to violins plots to add some info and some nice colors :
ggstatsplot::ggbetweenstats(ISLR::Wage, education, wage, "np") +
ylab("Workers raw wage") +
xlab("Education level")
1.4 Scatter plots
Scatter plots are not often used in my field of work, but they can come in handy in niche situations. I especially like the 3D scatterplot !
1.4.1 3D
# Prepare the data
dataScatter <- iris %>%
mutate(
# Assign colors to each species using hex color codes
p_col = ifelse(Species == "setosa", "#EB5E55",
ifelse(Species == "versicolor", "#C6D8D3", "#9984D4"))
)
# Initialize the 3D scatter plot
p <- plot_ly() %>%
add_trace(
data = dataScatter, # Use the prepared dataset
type = "scatter3d", # Create a 3D scatter plot
mode = "markers", # Display only markers (points)
x = ~Sepal.Length, # Map Sepal Length to the x-axis
y = ~Sepal.Width, # Map Sepal Width to the y-axis
z = ~Petal.Length, # Map Petal Length to the z-axis
color = ~Species, # Use Species as a grouping variable for the legend
marker = list(size = 5, color = ~p_col) # Set marker size and color based on p_col
) %>%
layout(
# Configure 3D axis titles
scene = list(
xaxis = list(title = 'Sepal length'),
yaxis = list(title = 'Sepal width'),
zaxis = list(title = 'Petal length')
),
# Add annotations (e.g., legend title) to the plot
annotations = list(
x = 1.13, # x-coordinate for the annotation
y = 1.05, # y-coordinate for the annotation
text = 'Species', # Annotation text
showarrow = FALSE # Hide any arrow for the annotation
)
)
# Add lines anchoring points to the XY plane
for (i in 1:nrow(dataScatter)) {
p <- p %>%
add_trace(
type = "scatter3d", # Add 3D scatter traces for lines
mode = "lines", # Render as lines
x = c(dataScatter[i, 1], dataScatter[i, 1]), # x-coordinates for the line (constant)
y = c(dataScatter[i, 2], dataScatter[i, 2]), # y-coordinates for the line (constant)
z = c(dataScatter[i, 3], min(dataScatter[, 3])), # z-coordinates: from point to minimum z-value
line = list(
# Dynamically generate RGBA color from p_col for line transparency
color = sprintf("rgba(%d, %d, %d, %.2f)",
grDevices::col2rgb(dataScatter$p_col[i])[1], # Red component
grDevices::col2rgb(dataScatter$p_col[i])[2], # Green component
grDevices::col2rgb(dataScatter$p_col[i])[3], # Blue component
0.8) # Transparency (alpha)
),
showlegend = FALSE # Exclude lines from the legend
)
}
# Display the final plot
p2 Regressions
This section is dedicated to regression models outputs and other associated graph.
2.1 Survival curve
Starting with the survival analysis, the most classic graph is the descending survival curve :
# Simulate survival data
n <- 1000 # Number of individuals
dataSurvCurve <- data.frame(
id = 1:n, # Unique identifier for each individual
time = as.numeric(sample(1:365, n, replace = TRUE)), # Simulated follow-up time (1 to 365 days)
event = factor(c(
sample(0:1, n/2, prob = c(0.4, 0.6), replace = TRUE), # Events for the unexposed group
sample(0:1, n/2, prob = c(0.7, 0.3), replace = TRUE) # Events for the exposed group
)), # Event status as a factor: 0 = censored, 1 = event
exposure = rep(c("Unexposed", "Exposed"), each = n/2) # Exposure groups
)
# Adjust time for censored observations (event == 0)
dataSurvCurve$time[dataSurvCurve$event == 0] <- 365
# Fit the survival model with event stratified by exposure
fit <- survfit(Surv(time = time, event = event) ~ exposure, data = dataSurvCurve)
# Define the upper limit of the y-axis in percentage
ymax <- 30
# Plot survival curves using autoplot from the ggplot2 framework
autoplot(fit, surv.linetype = 'blank', surv.colour = NA) + # Start with a blank base plot
geom_step(aes(linetype = strata, color = event), size = 0.8) + # Add step lines by strata and event
scale_linetype_manual(
name = "Group", # Legend title for group
values = c('dotted', 'solid'), # Line types for the groups
labels = c('Non-exposed', 'Exposed') # Labels for the groups
) +
scale_color_manual(
name = "Event", # Legend title for events
values = c("white", "#1C4073"), # Colors for censored and events
labels = c(' ', "Hospitalised") # Labels for the events
) +
labs(
x = "Follow up period (days)", # X-axis label (French: Follow-up period in days)
y = "Proportion of population", # Y-axis label
title = "" # Empty title
) +
geom_vline(aes(xintercept = 0), size = 1) + # Add vertical line at x = 0
geom_hline(aes(yintercept = 1 - ymax / 100), size = 1) + # Add horizontal line for the ymax threshold
scale_y_reverse(
limits = 1 - c(ymax / 100, 1), # Reverse y-axis to represent decreasing survival
breaks = 1 - seq(ymax / 100, 1, 0.1), # Break points for y-axis
labels = paste0(seq(ymax, 100, 10), "%"), # Labels as percentages
expand = c(0, 0) # No expansion around the limits
) +
scale_x_continuous(
breaks = seq(0, 360, 60), # X-axis breaks every 60 days
expand = c(0, 0) # No expansion around the limits
) +
theme_minimal() + # Use a minimal theme
theme(
plot.title = element_text(hjust = 0.5), # Center align the title
axis.title.x = element_text(face = "bold", colour = "black", size = 12), # Style x-axis title
axis.title.y = element_text(face = "bold", colour = "black", size = 12), # Style y-axis title
legend.title = element_text(face = "bold", size = 10), # Style legend title
panel.grid.major.y = element_line(colour = "lightgray", size = 0.3), # Grid for y-axis
panel.grid.minor.x = element_blank(), # Remove minor grid for x-axis
legend.position = c(0.15, 0.30), # Position legend inside plot
legend.background = element_rect(fill = "transparent", color = "white") # Transparent background for legend
)
Second version of almost the same curve, adding some confidence interval using ggsurvplot() :
# Simulate survival data for analysis
n <- 1000 # Number of individuals
dataSurvCurve <- data.frame(
id = 1:n, # Unique identifier for each individual
time = as.numeric(sample(1:365, n, replace = TRUE)), # Simulated follow-up time between 1 and 365 days
event = c(
sample(0:1, n/2, prob = c(0.4, 0.6), replace = TRUE), # First group: 40% censored, 60% events
sample(0:1, n/2, prob = c(0.7, 0.3), replace = TRUE) # Second group: 70% censored, 30% events
),
# Event status: 0 = censored, 1 = event occurred
exposure = rep(c("Unexposed", "Exposed"), each = n/2) # Exposure status: "Unexposed" or "Exposed"
)
# Fit survival curves stratified by exposure
fit <- survfit(Surv(time = time, event = event) ~ exposure, data = dataSurvCurve)
# Plot the survival curves using ggsurvplot
ggsurvplot(
fit, # Survival fit object
data = dataSurvCurve, # Data source
legend.title = "Group", # Title for the legend
legend.labs = c("Exposed", "Unexposed"), # Labels for legend groups
conf.int = TRUE, # Display confidence intervals
censor = FALSE, # Do not display censoring marks
xlab = "Follow up time (days)", # Label for x-axis
ylab = "Non-hospitalisation probability", # Label for y-axis
break.time.by = 60, # Breaks on the x-axis every 60 days
surv.scale = "percent", # Scale survival probability as percentages
ylim = c(0, 1), # Set y-axis limits (0% to 100%)
axes.offset = FALSE, # Align axes at the origin
legend = c(0.12, 0.15) # Position the legend within the plot
)
Let’s add one more event and have an some components (I would use ggsurvplot(), but it does not handle a survfit object having a non binary event) :
# Simulating survival data
n <- 1000 # Number of individuals
dataSurvCurve <- data.frame(
id = 1:n,
time = as.numeric(sample(1:365, n, replace = TRUE)), # Simulated follow-up time between 1 and 365 days
event = factor(c(sample(0:2, n/2, prob = c(0.4, 0.4, 0.2), replace = TRUE),
sample(0:2, n/2, prob = c(0.2, 0.3, 0.5), replace = TRUE))), # Event (0 = censored, 1 = sick, 2 = dead)
exposure = rep(c("Unexposed", "Exposed"), each = n/2) # Exposure status (Unexposed or Exposed)
)
dataSurvCurve$time[dataSurvCurve$event == 0] <- 365 # Set time to 365 for censored observations (event == 0)
fit <- survfit(Surv(time = time, event = event) ~ exposure, data = dataSurvCurve) # Fitting a Survival Model
# Plot a survival curve with `autoplot` and enhance it with additional layers
autoplot(fit, surv.linetype = 'blank', surv.colour = NA) +
# Add step lines representing survival curves
geom_step(
aes(linetype = strata, color = event), # Map line type to `strata` and color to `event`
size = 0.8 # Set line thickness
) +
# Customize the line types for survival curves
scale_linetype_manual(
name = "Group", # Title for the legend
values = c('dashed', 'solid'), # Define line types: dashed for unexposed, solid for exposed
labels = c('Unexposed', 'Exposed') # Labels for the legend
) +
# Customize the colors for the event states
scale_color_manual(
name = "State", # Title for the legend
values = c("#AAC0AF", "#1C4073", "#B28B84"), # Colors for Healthy, Sick, and Dead states
labels = c('Healthy', 'Sick', 'Dead') # Labels for the legend
) +
# Add axis labels and a title
labs(
x = "Followup period (days)", # X-axis label
y = "Proportion of the population", # Y-axis label
title = "" # Title (empty for now)
) +
# Add a vertical reference line at x = 0
geom_vline(
aes(xintercept = 0), # Position of the line
size = 1 # Thickness of the line
) +
# Customize the y-axis scale
scale_y_continuous(
limits = c(0, 1), # Set the y-axis range from 0 to 1
breaks = seq(0, 1, 0.1), # Major ticks every 0.1
labels = paste0(seq(0, 100, 10), "%"), # Convert tick labels to percentages
expand = c(0, 0) # Remove padding on the y-axis
) +
# Customize the x-axis scale
scale_x_continuous(
breaks = seq(0, 360, 60), # Major ticks every 60 days
expand = c(0, 0) # Remove padding on the x-axis
) +
# Apply a minimal theme for a clean appearance
theme_minimal() +
# Customize specific theme elements
theme(
plot.title = element_text(hjust = 0.5), # Center-align the plot title
axis.title.x = element_text(
face = "bold", # Make x-axis title bold
colour = "black", # Set title color to black
size = 12 # Set font size
),
axis.title.y = element_text(
face = "bold", # Make y-axis title bold
colour = "black", # Set title color to black
size = 12 # Set font size
),
legend.title = element_text(face = "bold", size = 10), # Style legend titles
panel.grid.major.y = element_line(
colour = "lightgray", # Light gray major grid lines on the y-axis
size = 0.3 # Set line thickness
),
panel.grid.minor.x = element_blank() # Remove minor grid lines on the x-axis
)
Lastly, combining different curves using ggsurvplot() :
# Create a data frame for survival analysis
dataSurvCurve <- data.frame(
os.time = colon$time, # Overall survival time from 'colon' dataset
os.status = colon$status, # Overall survival event status (1 = event, 0 = censored)
pfs.time = sample(colon$time), # Progression-free survival time, randomized for illustration
pfs.status = colon$status, # Progression-free survival event status (1 = event, 0 = censored)
sex = colon$sex, # Patient sex
rx = colon$rx, # Treatment type
adhere = colon$adhere # Adherence to treatment
)
# Fit survival curves for progression-free survival (PFS) by treatment type
fit.pfs <- survfit(Surv(pfs.time, pfs.status) ~ rx, data = dataSurvCurve)
# Fit survival curves for overall survival (OS) by treatment type
fit.os <- survfit(Surv(os.time, os.status) ~ rx, data = dataSurvCurve)
# Combine the survival fits into a list for plotting
fits <- list(PFS = fit.pfs, OS = fit.os)
# Plot the survival curves using ggsurvplot
ggsurvplot(
fits, # List of survival fits (PFS and OS)
data = dataSurvCurve, # Data used for plotting
combine = TRUE, # Combine PFS and OS plots into one
censor = FALSE, # Do not display censoring marks
palette = "jco", # Use JCO color palette
surv.scale = "percent", # Display survival probabilities as percentages
ylim = c(0, 1), # Set y-axis limits (0% to 100% survival)
axes.offset = FALSE, # Align axes at the origin
legend = "right", # Position the legend on the right
xlab = "Survival time (years)",# Label for x-axis
xscale = "d_y", # Scale x-axis from days to years
break.x.by = 365.25, # Breaks on the x-axis every year
legend.title = "Treatment type" # Title for the legend
)
Or display the different cruves in a grid :
fit <- survfit( Surv(time, status) ~ sex, data = colon)
ggsurvplot_facet(fit, colon, facet.by = "rx",
palette = "jco", # Combine PFS and OS plots into one
censor = FALSE, # Use JCO color palette
surv.scale = "percent", # Display survival probabilities as percentages
ylim = c(0.3, 1), # Set y-axis limits (0% to 100% survival)
axes.offset = FALSE, # Align axes at the origin
legend = c("PFS", "OS"), # Position the legend on the right
xlab = "Survival time (years)",# Label for x-axis
xscale = "d_y", # Scale x-axis from days to years
break.x.by = 365.25, # Breaks on the x-axis every year
legend.title = "Treatment type", # Title for the legend)
conf.int = TRUE
)
2.2 Forest plots
Forest plots are the go to graph for any type of regression that present results as exponential of it’s coefficients. I really like the existing packages displaying the OR and CI95% so I tried to reproduce it.
# Create the data frame for the forest plot
dataForest <- data.frame(
var = c("Age", "Age", "Age", "Age", "Sex", "Sex", "Diabetes", "Diabetes", "Smoker", "Smoker"), # Variables
mod = c("18-24", "25-44", "45-64", "64+", "Male", "Female", "No", "Yes", "No", "Yes"), # Categories or modalities within variables
estimate = c(NA, 1.05, 1.38, 1.72, NA, 0.87, NA, 2.48, NA, 0.97), # Point estimates (Odds Ratios)
conf.low = c(NA, 0.92, 1.21, 1.44, NA, 0.72, NA, 2.13, NA, 0.88), # Lower bounds of confidence intervals
conf.high = c(NA, 1.22, 1.63, 1.87, NA, 1.01, NA, 2.67, NA, 1.05) # Upper bounds of confidence intervals
) %>%
mutate(
id = 1:10, # Assign a unique ID to each row
label = ifelse(is.na(estimate), "Ref",
paste0(formatC(estimate, 2, format = "f"), " [",
formatC(conf.low, 2, format = "f"), "-",
formatC(conf.high, 2, format = "f"), "]")),
# Create labels for display: "estimate [conf.low-conf.high]" or "Ref" if missing
estimate = ifelse(is.na(estimate), 1, estimate), # Replace missing estimates with 1 (reference)
conf.low = ifelse(is.na(conf.low), 1, conf.low), # Replace missing lower bounds with 1 (reference)
conf.high = ifelse(is.na(conf.high), 1, conf.high), # Replace missing upper bounds with 1 (reference)
signif = ifelse(conf.low <= 1 & conf.high >= 1, "p>0.05", "p<0.05")
# Determine significance: overlaps 1 (p>0.05) or not (p<0.05)
)
# Create the forest plot using ggplot2
ggplot(dataForest) +
geom_vline(
xintercept = 1, # Reference line at Odds Ratio = 1
color = "gray25",
linetype = 2 # Dashed line
) +
geom_vline(
xintercept = c(0.25, 0.5, 2, 4), # Additional vertical lines for log-scale guidance
color = "gray75",
linetype = "dotted"
) +
geom_errorbar(
aes(
y = reorder(mod, -id), # Reorder y-axis labels based on ID
xmin = conf.low, # Lower bound of the confidence interval
xmax = conf.high # Upper bound of the confidence interval
),
position = position_dodge(0.5), # Adjust error bar position
width = 0.4, # Width of the error bars
size = 0.8 # Thickness of the error bars
) +
geom_point(
aes(x = estimate, y = reorder(mod, -id), color = signif), # Add points for the estimates
position = position_dodge(width = 0.5), # Adjust point position
size = 2.5 # Size of the points
) +
scale_x_log10( # Use a logarithmic scale for x-axis
breaks = c(0.5, 1, 2, 4), # Specify tick marks
limits = c(0.49, 6) # Set x-axis limits
) +
scale_color_manual(
name = "Significance", # Legend title for significance
values = c("red", "black") # Colors for significant and non-significant points
) +
facet_grid(
rows = vars(var), # Create separate panels for each variable
scales = "free", # Allow scales to adjust freely per panel
switch = "both" # Switch facet labels to both sides
) +
xlab("Odds ratio") + # Label for the x-axis
ylab("") + # Remove y-axis label
theme_classic() + # Use a clean, minimal theme
theme(panel.grid.major.y = element_line(colour = "lightgray")) + # Add gridlines for readability
geom_text(
aes(x = 5, y = reorder(mod, -id), label = dataForest$label), # Add text labels for estimates
)
Here is a simpler version, with no OR table and a lighter display :
# Create the data frame for the forest plot
dataForest <- data.frame(
var = c("Age", "Age", "Age", "Age", "Sex", "Sex", "Diabetes", "Diabetes", "Smoker", "Smoker"), # Variables
mod = c("18-24", "25-44", "45-64", "64+", "Male", "Female", "No", "Yes", "No", "Yes"), # Categories or modalities within variables
estimate = c(NA, 1.05, 1.38, 1.72, NA, 0.87, NA, 2.48, NA, 0.97), # Point estimates (Odds Ratios)
conf.low = c(NA, 0.92, 1.21, 1.44, NA, 0.72, NA, 2.13, NA, 0.88), # Lower bounds of confidence intervals
conf.high = c(NA, 1.22, 1.63, 1.87, NA, 1.01, NA, 2.67, NA, 1.05) # Upper bounds of confidence intervals
) %>%
dplyr::filter(!is.na(estimate)) %>%
mutate(
id = 1:6, # Assign a unique ID to each row
mod = paste0(var, " = ", mod),# Create labels for display
estimate = ifelse(is.na(estimate), 1, estimate), # Replace missing estimates with 1 (reference)
conf.low = ifelse(is.na(conf.low), 1, conf.low), # Replace missing lower bounds with 1 (reference)
conf.high = ifelse(is.na(conf.high), 1, conf.high), # Replace missing upper bounds with 1 (reference)
signif = ifelse(conf.low <= 1 & conf.high >= 1, "p>0.05", "p<0.05")
# Determine significance: overlaps 1 (p>0.05) or not (p<0.05)
)
# Create the forest plot using ggplot2
ggplot(dataForest) +
geom_vline(
xintercept = 1, # Reference line at Odds Ratio = 1
color = "gray25",
linetype = 2 # Dashed line
) +
geom_vline(
xintercept = c(0.25, 0.5, 2, 4), # Additional vertical lines for log-scale guidance
color = "gray75",
linetype = "dashed"
) +
geom_errorbar(
aes(
y = reorder(mod, -id), # Reorder y-axis labels based on ID
xmin = conf.low, # Lower bound of the confidence interval
xmax = conf.high # Upper bound of the confidence interval
),
position = position_dodge(0.5), # Adjust error bar position
width = 0.2, # Width of the error bars
size = 0.8 # Thickness of the error bars
) +
geom_point(
aes(x = estimate, y = reorder(mod, -id), shape = signif, size = signif), # Add points for the estimates
position = position_dodge(width = 0.5), # Adjust point position
) +
scale_x_log10( # Use a logarithmic scale for x-axis
breaks = c(0.5, 1, 2, 4), # Specify tick marks
limits = c(0.49, 6) # Set x-axis limits
) +
scale_shape_manual(
name = "Significance", # Legend title for significance
values = c(15, 18) # Colors for significant and non-significant points
) +
scale_size_manual(
name = "Significance", # Legend title for significance
values = c(4, 2) # Colors for significant and non-significant points
) +
xlab("Odds ratio") + # Label for the x-axis
ylab("") + # Remove y-axis label
theme_classic() + # Use a clean, minimal theme
theme(panel.grid.major.y = element_line(colour = "lightgray", linetype = "dotted")) # Add gridlines for readability
3 Epidemiology
This section is dedicated to graphs that are often used in epidemiology. Those graph can of course be used in other fields (data don’t care about your feelings), but I am obviously speaking in the context of my own job. I am indeed biased.
3.1 Epidemic curve
Epidemic curves are just like bar plots, but the time component displayed on the X axis make it so that we tend to represent it more like and histogram.
Here, I am displaying the two data series as dodged curves, and adding a second axis to displayed the line :
dataCurve <- data.frame(
Incident = c(rep("Severe Covid-19 case", 464), rep("Severe adverse reaction", 76)),
Date = rep(seq.POSIXt(as_datetime("2020-06-01 00:00:00", tz = "CET"),
as_datetime("2023-02-01 00:00:00", tz = "CET"), by = "month"),
c(19, 11, 11, 9, 13, 21, 58, 49, 73, 83, 33, 19, 14 ,13, 16, 11,
3, 3, 11, 26, 8, 2, 4, 3, 8, 1, 4, 2, 4, 4, 2, 1, 1))
)
dataDoses <- data.frame(
Date = seq.Date(as.Date("2021-01-01"), as.Date("2023-01-01"), by = "month"),
Ndoses = c(4573, 5343, 5474, 27699, 73329, 137575, 109853, 62694, 29315, 9604,
10456, 95054, 81274, 17859, 3998, 2385, 2384, 1772, 1931, 1471,
2326, 1872, 7005, 9746, 4759)
)
monthly_breaks <- seq.POSIXt(
from = as_datetime("2020-01-01 00:00:00", tz = "CET"),
to = as_datetime("2023-04-01 00:00:00", tz = "CET"),
by = "month"
)
ymax <- 87.5
coeff <- (140000 / ymax)
dataDoses$Ndoses <- dataDoses$Ndoses / coeff
# Initialize the plot with a dataset (DATA_ex1_1)
ggplot(dataCurve) +
# Add a histogram with dodged bars, colored by the "Incident" variable
geom_histogram(
mapping = aes(x = Date, fill = Incident), # Map 'cas_date' to x-axis, and 'Incident' to fill
breaks = monthly_breaks, # Use specified monthly breaks
closed = "left", # Define intervals as left-closed
color = "white", # Set white border for bars
position = "dodge" # Dodge bars side by side
) +
# Add horizontal grid lines with white color
geom_hline(yintercept = 1:ymax, color = "white") +
# Add vertical reference lines for specific dates
geom_vline(
xintercept = as_datetime(c("2021-01-01 00:00:00", "2022-01-01 00:00:00"), tz = "CET"), # Define dates as datetime objects
color = "lavenderblush4", # Set line color
size = 1 # Set line thickness
) +
# Annotate rectangular periods of interest
annotate(
"rect", # Specify rectangle annotation
xmin = as_datetime("2020-04-01 00:00:00", tz = "CET"), # Start of the rectangle
xmax = as_datetime("2020-06-01 00:00:00", tz = "CET"), # End of the rectangle
ymin = 0, # Rectangle starts at y = 0
ymax = ymax, # Rectangle ends at y = ymax
fill = "gray", # Set fill color to gray
alpha = 0.25 # Set transparency
) +
# Repeat annotation for additional periods
annotate("rect", xmin = as_datetime("2020-11-01 00:00:00", tz = "CET"),
xmax = as_datetime("2021-01-01 00:00:00", tz = "CET"), ymin = 0, ymax = ymax, fill = "gray", alpha = 0.25) +
annotate("rect", xmin = as_datetime("2021-04-01 00:00:00", tz = "CET"),
xmax = as_datetime("2021-06-01 00:00:00", tz = "CET"), ymin = 0, ymax = ymax, fill = "gray", alpha = 0.25) +
# Add a line plot for data in DATA_ex1_2
geom_line(
data = dataDoses,
aes(
x = as_datetime(paste(Date, "00:00:00"), tz = "CET"), # Convert Date to datetime
y = Ndoses, # Use Ndoses for y-axis
colour = "Number of vaccine doses" # Set color legend
),
linetype = 5 # Set line type
) +
# Add points on the line plot
geom_point(
data = dataDoses,
aes(x = as_datetime(paste(Date, "00:00:00"), tz = "CET"), y = Ndoses), # Map Date and Ndoses
colour = "darkslateblue", # Set color
shape = 16, # Use solid circles
size = 2 # Set point size
) +
# Format x-axis as datetime
scale_x_datetime(
expand = c(0, 0), # Remove padding
date_breaks = "month", # Major breaks every month
date_minor_breaks = "month", # Minor breaks also monthly
date_labels = "%m/%y" # Format as MM/YY
) +
# Customize color scale for doses
scale_colour_manual("Doses", values = "darkslateblue") +
# Use classic theme for the plot
theme_classic() +
# Modify various theme elements
theme(
panel.grid.minor.y = element_line(colour = "lightgray", size = 1), # Minor grid lines
panel.grid.major.y = element_line(colour = "lavenderblush4", size = 1), # Major grid lines
axis.text.x = element_text(angle = 90), # Rotate x-axis labels
axis.text.y.right = element_text(color = "darkslateblue"), # Color right y-axis text
axis.line.y.right = element_line(color = "darkslateblue"), # Color right y-axis line
axis.title.y.right = element_text(color = "darkslateblue"), # Color right y-axis title
legend.position = "bottom"
) +
# Format y-axis with a secondary axis
scale_y_continuous(
expand = c(0, 0), # Remove padding
limits = c(0, ymax), # Set y-axis limits
breaks = seq(0, 90, 10), # Major breaks every 10
sec.axis = sec_axis(
trans = ~ . * coeff, # Transform secondary axis
name = "Number of doses dispensed", # Secondary axis title
breaks = seq(0, 90, 10)*coeff, # Secondary axis breaks
labels = formatC( # Format secondary axis labels
seq(0, 90, 10)*coeff,
format = "f",
big.mark = " ",
digits = 0
)
)
) +
# Add titles and axis labels
labs(title = "", x = " ", y = "Number of incidents") +
# Add text annotations for specific periods
geom_text(mapping = aes_q(
x = as_datetime("2020-07-01 00:00:00", tz = "CET"), y = 82, label = "Period n°1"
)) +
geom_text(mapping = aes_q(
x = as_datetime("2021-07-01 00:00:00", tz = "CET"), y = 82, label = "Period n°2"
)) +
geom_text(mapping = aes_q(
x = as_datetime("2022-08-01 00:00:00", tz = "CET"), y = 82, label = "Period n°3"
))
It is easier to compare the shape of several curves when they are displayed separately :
# Create a data frame for histogram visualization
dataCurve <- data.frame(
# "Sex" column with repeated values for "Male" and "Female"
Sex = c(rep("Male", 323), rep("Female", 349)),
# "Date" column with monthly timestamps repeated according to specified counts
Date = c(
# Male dates: repeated monthly sequence with specific counts for each month
rep(seq.POSIXt(
as_datetime("2024-01-01 00:00:00", tz = "CET"),
as_datetime("2025-01-01 00:00:00", tz = "CET"),
by = "month"
), c(15, 21, 27, 26, 33, 38, 30, 22, 28, 17, 20, 19, 27)),
# Female dates: repeated monthly sequence with specific counts for each month
rep(seq.POSIXt(
as_datetime("2024-01-01 00:00:00", tz = "CET"),
as_datetime("2025-01-01 00:00:00", tz = "CET"),
by = "month"
), c(34, 41, 56, 32, 29, 33, 37, 26, 21, 11, 9, 12, 8))
)
)
# Define monthly breaks for the histogram
monthly_breaks <- seq.POSIXt(
from = as_datetime("2024-01-01 00:00:00", tz = "CET"), # Start of the range
to = as_datetime("2025-01-01 00:00:00", tz = "CET"), # End of the range
by = "month" # Break intervals (monthly)
)
# Create the histogram using ggplot2
ggplot(dataCurve) +
# Add a histogram layer
geom_histogram(
mapping = aes(x = Date), # Map the "Date" column to the x-axis
breaks = monthly_breaks, # Use predefined monthly breaks for bins
closed = "left", # Bins include the left boundary
color = "white", # White borders around bars
position = "dodge", # Position bars side by side
fill = "#0f766e" # Fill color for bars
) +
# Add horizontal lines at intervals of 5
geom_hline(
yintercept = seq(5, ymax, by = 5), # Sequence of horizontal line positions
color = "white" # Line color
) +
# Add a horizontal line at y=0
geom_hline(
aes(yintercept = 0)
) +
# Customize x-axis labels and ticks
scale_x_datetime(
expand = c(0, 0), # No expansion at the axis limits
date_breaks = "month", # Major breaks at each month
date_minor_breaks = "month", # Minor breaks at each month
date_labels = "%m/%y" # Format labels as "month/year"
) +
# Customize y-axis labels and ticks
scale_y_continuous(
expand = c(0, 0), # No expansion at the axis limits
limits = c(0, 60), # Set y-axis limits between 0 and 60
breaks = seq(0, 60, 10) # Major breaks at intervals of 10
) +
# Add plot titles and axis labels
labs(
title = "", # Main title (empty)
caption = "", # Caption (empty)
x = " ", # X-axis label (empty space)
y = "Number of cases" # Y-axis label
) +
# Create separate panels for each sex
facet_grid(
rows = vars(Sex) # Facet by "Sex" variable (one row for each value)
) +
# Apply a classic theme
theme_classic() +
# Customize the appearance of grid lines and axis text
theme(
panel.grid.minor.y = element_line(
colour = "lightgray", # Minor grid line color
size = 1 # Minor grid line size
),
panel.grid.major.y = element_line(
colour = "lavenderblush4", # Major grid line color
size = 1 # Major grid line size
)
)
Sometimes, you want to now the total number of case, but still display the factor variable :
# Create a data frame for the bar plot
dataCurve <- data.frame(
# "Sex" column with repeated values for "Male" and "Female"
Sex = c(rep("Male", 323), rep("Female", 349)),
# "Date" column with monthly timestamps repeated according to specified counts
Date = c(
# Male dates: repeated monthly sequence with specific counts for each month
rep(seq.POSIXt(
as_datetime("2024-01-01 00:00:00", tz = "CET"),
as_datetime("2025-01-01 00:00:00", tz = "CET"),
by = "month"
), c(15, 21, 27, 26, 33, 38, 30, 22, 28, 17, 20, 19, 27)),
# Female dates: repeated monthly sequence with specific counts for each month
rep(seq.POSIXt(
as_datetime("2024-01-01 00:00:00", tz = "CET"),
as_datetime("2025-01-01 00:00:00", tz = "CET"),
by = "month"
), c(34, 41, 56, 32, 29, 33, 37, 26, 21, 11, 9, 12, 8))
)
)
# Define monthly breaks for the bar plot
monthly_breaks <- seq.POSIXt(
from = as_datetime("2024-01-01 00:00:00", tz = "CET"), # Start of the range
to = as_datetime("2025-01-01 00:00:00", tz = "CET"), # End of the range
by = "month" # Break intervals (monthly)
)
# Create the bar plot using ggplot2
ggplot(dataCurve) +
# Add a stacked bar plot layer
geom_bar(
mapping = aes(
x = Date, # Map "Date" to the x-axis
fill = Sex # Use "Sex" for the fill aesthetic to create stacked bars
),
breaks = monthly_breaks, # Define bin breaks using the monthly breaks
closed = "left", # Bins include the left boundary
color = "white", # White borders around bars
position = "stack" # Stack the bars by "Sex"
) +
# Add horizontal lines at regular intervals
geom_hline(
yintercept = seq(5, ymax, by = 5), # Sequence of horizontal line positions
color = "white" # Line color
) +
# Add a horizontal line at y=0
geom_hline(
aes(yintercept = 0)
) +
# Customize x-axis labels and ticks
scale_x_datetime(
expand = c(0, 0), # No expansion at the axis limits
date_breaks = "month", # Major breaks at each month
date_minor_breaks = "month", # Minor breaks at each month
date_labels = "%m/%y" # Format labels as "month/year"
) +
# Customize y-axis labels and ticks
scale_y_continuous(
expand = c(0, 0), # No expansion at the axis limits
limits = c(0, 90), # Set y-axis limits between 0 and 90
breaks = seq(0, 90, 10) # Major breaks at intervals of 10
) +
# Add plot titles and axis labels
labs(
title = "", # Main title (empty)
caption = "", # Caption (empty)
x = " ", # X-axis label (empty space)
y = "Number of cases" # Y-axis label
) +
# Apply a classic theme
theme_classic() +
# Customize the appearance of grid lines
theme(
panel.grid.minor.y = element_line(
colour = "lightgray", # Minor grid line color
size = 1 # Minor grid line size
),
panel.grid.major.y = element_line(
colour = "lavenderblush4", # Major grid line color
size = 1 # Major grid line size
)
)
3.2 Care trajectories
Care trajectories are a huge deal in epidemiology. Succession of treatment states can be tricky to display, and is it definitly an pain to handle the data to mke it work.
3.2.1 Sankey
Sankey diagrams are cool. I like them. Look at it :
createSankeyData <- function(data, states, timesColumns) {
# Function to prepare data for a Sankey diagram
# Arguments:
# data: A dataframe containing sequential data
# states: A vector of unique states
# timesColumns: A vector of column names representing time steps
data <- as.data.frame(data) # Ensure the input is a dataframe
n_states <- length(states) # Number of unique states
n_times <- length(timesColumns) # Number of time columns (steps)
# Convert time columns into factors with specified levels (states)
for (i in 1:n_times) {
data[, timesColumns[i]] <- factor(data[, timesColumns[i]], levels = states)
}
# Define colors for the nodes (states) and links between states
nodesCols <- c("#AAC0AF", "#B28B84", "#1C4073", "#0f766e", "#653239", "#472C1B", "#5C2751")[1:n_states]
linksCols <- c("#D0DCD3", "#D0B8B4", "#285CA4", "#17B5A7", "#964A54", "#76492D", "#8F3D7E")[1:n_states]
vals <- c() # Initialize a vector to store transition counts
# Calculate transitions between consecutive time steps for each state
for (i in 2:n_times) {
for (j in 1:n_states) {
# Count occurrences of transitions from state j at time (i-1) to all states at time i
vals <- c(vals, table(data[, timesColumns[i]][data[, timesColumns[i - 1]] == states[j]]))
}
}
# Prepare Sankey diagram data structure
dataSankeyTem <- list(
Nodes = data.frame(
X = 1:(n_states * n_times), # Node indices (unique for each state and time step)
label = rep(states, n_times), # Node labels (repeated for each time step)
color = rep(nodesCols, n_times) # Node colors
),
Links = data.frame(
source = c(rep(1:(n_states * (n_times - 1)), each = n_states)) - 1, # Source nodes for transitions
target = as.vector(sapply(split((n_states + 1):(n_states * n_times),
rep(1:(n_times - 1), each = n_states)),
function(x) {rep(x, n_states)})) - 1, # Target nodes for transitions
value = vals, # Transition counts
color = rep(rep(linksCols, each = n_states), n_times - 1) # Colors for links
)
)
}
# Example usage: Create Sankey data for three time steps with predefined states
sankeyData <- createSankeyData(
data.frame(
T1 = sample(
c("Initial treatment", "Substitution", "No treatment"),
size = 1000, prob = c(0.95, 0.025, 0.025), replace = TRUE),
T2 = sample(
c("Initial treatment", "Substitution", "No treatment"),
size = 1000, prob = c(0.75, 0.125, 0.125), replace = TRUE),
T3 = sample(
c("Initial treatment", "Substitution", "No treatment"),
size = 1000, prob = c(0.5, 0.25, 0.25), replace = TRUE)),
c("Initial treatment", "Substitution", "No treatment"), # Define states
c("T1", "T2", "T3") # Define time columns
)
plotSankey <- function(Nodes, Links, add.prop = FALSE) {
# Function to create a Sankey diagram using plotly.
# Args:
# Nodes: Data frame containing node details (labels, colors, etc.).
# Links: Data frame containing link details (source, target, values, colors).
# add.prop: Logical. If TRUE, adds percentage proportions to node labels.
if (add.prop) {
# Check if all sources in Links have an equal number of occurrences
if (all(table(Links$source) == table(Links$source)[1])) {
nrow_per_times <- table(Links$source)[1] ^ 2 # Determine rows per iteration
n_times <- nrow(Links) / nrow_per_times # Calculate number of iterations
for (i in 1:n_times) {
# Extract subset of Links for current iteration
tab <- Links[((i - 1) * nrow_per_times + 1):((i - 1) * nrow_per_times + nrow_per_times), ]
# Calculate proportions for targets
vals <- as.numeric(by(tab$value, tab$target, sum)) # Sum values by target
Nodes[Nodes$X %in% (unique(tab$target) + 1), "label"] <- paste0(
Nodes[Nodes$X %in% (unique(tab$target) + 1), "label"],
" (",
formatC(vals / sum(vals) * 100, digits = 1, format = 'f'), # Format as percentage
"%)"
)
if (i == 1) {
# Calculate proportions for sources during the first iteration
vals <- as.numeric(by(tab$value, tab$source, sum)) # Sum values by source
Nodes[Nodes$X %in% (unique(tab$source) + 1), "label"] <- paste0(
Nodes[Nodes$X %in% (unique(tab$source) + 1), "label"],
" (",
formatC(vals / sum(vals) * 100, digits = 1, format = 'f'), # Format as percentage
"%)"
)
}
}
} else {
warning(
"Links arg does not have equal number of each source and function cannot automatically calculate proportions."
)
}
}
# Convert colors in Links to RGBA format with transparency
Links$color <- apply(grDevices::col2rgb(Links$color), 2, function(x) {
paste0("rgba(", x[1], ",", x[2], ",", x[3], ",0.4)") # Add 40% transparency
})
# Create the Sankey diagram using plotly
fig <- plot_ly(
type = "sankey", # Specify Sankey diagram
orientation = "h", # Horizontal orientation
alpha_stroke = 0.2, # Node border transparency
node = list(
label = Nodes$label, # Node labels
color = Nodes$color, # Node colors
pad = 15, # Padding between nodes
thickness = 20, # Node thickness
line = list(color = "black", width = 0.5) # Node border style
),
link = list(
source = Links$source, # Source nodes for links
target = Links$target, # Target nodes for links
value = Links$value, # Link values
color = Links$color # Link colors (RGBA)
)
)
# Customize the layout of the Sankey diagram
fig <- fig %>% layout(
font = list(
size = 14, # Font size for labels
color = "black", # Font color
weight = "bold" # Font weight
)
)
# Return the generated plot
fig
}
plotSankey(sankeyData$Nodes, sankeyData$Links, add.prop = TRUE)3.2.2 Carpet
Carpet diagrams allow to display some individual trajectories. I really enjoy the facts that you can display the whole population and/or the clusters identified via a clustering method.
# Create a dataset simulating treatment sequences
carpetData <- data.frame(
T1 = sample( # Initial treatment stage
c("Initial treatment", "Substitution", "No treatment"),
size = 100000, prob = c(0.95, 0.025, 0.025), replace = TRUE
),
T2 = sample( # Second treatment stage
c("Initial treatment", "Substitution", "No treatment"),
size = 100000, prob = c(0.75, 0.125, 0.125), replace = TRUE
),
T3 = sample( # Third treatment stage
c("Initial treatment", "Substitution", "No treatment"),
size = 100000, prob = c(0.65, 0.25, 0.1), replace = TRUE
),
T4 = sample( # Third treatment stage
c("Initial treatment", "Substitution", "No treatment"),
size = 100000, prob = c(0.5, 0.25, 0.25), replace = TRUE
),
T5 = sample( # Third treatment stage
c("Initial treatment", "Substitution", "No treatment"),
size = 100000, prob = c(0.35, 0.4, 0.25), replace = TRUE
)
) %>%
group_by(T1, T2, T3, T4, T5) %>% # Group data by unique sequences
summarise(w = n()) # Summarize the weight (frequency) of each sequence
# Define a sequence object using the TraMineR package
seqCarpet <- seqdef(
data = carpetData[, c("T1", "T2", "T3", "T4", "T5")], # Extract sequence columns
weights = carpetData$w, # Use weights for the sequences
cpal = c("#AAC0AF", "#B28B84", "#1C4073"), # Custom color palette for states
cnames = c("T1", "T2", "T3", "T4", "T5") # Custom names for the sequence stages
)
# Define substitution costs using a constant method
couts <- seqsubm(seqCarpet, method = "CONSTANT", cval = 2)
# Compute optimal matching distances
seq.om <- seqdist(seqCarpet,
method = "OM", # Optimal Matching (OM) method
indel = 1, # Insertion/deletion cost
sm = couts) # Substitution matrix
# Perform hierarchical clustering on the sequence distances
seq.dist <- hclust(as.dist(seq.om), method = "ward.D2")
# Create sequence clusters by cutting the dendrogram into 2 groups
seq_cl <- factor(cutree(seq.dist, 3), labels = c("Class n°1", "Class n°2", "Class n°3"))
# Plot individual sequence plots and group sequence plots side by side
ggarrange(
ggseqiplot(seqCarpet, facet_ncol = 1, no.n = TRUE) + # Individual plot
theme(
panel.spacing = unit(0.1, "lines"), # Adjust panel spacing
axis.text.y = element_text(colour = "white"), # Hide y-axis text
axis.ticks.y = element_line(colour = "white") # Hide y-axis ticks
) +
labs(y = ""), # Remove y-axis label
ggseqiplot( # Plot grouped sequences
seqCarpet,
group = seq_cl, # Group by clusters
facet_ncol = 1,
no.n = TRUE
) +
force_panelsizes(rows = table(seq_cl)) + # Adjust panel sizes by group
theme(
panel.spacing = unit(0.1, "lines"), # Adjust panel spacing
axis.text.y = element_text(colour = "white"), # Hide y-axis text
axis.ticks.y = element_line(colour = "white") # Hide y-axis ticks
) +
labs(y = ""), # Remove y-axis label
ncol = 2, # Arrange plots in 2 columns
nrow = 1, # Arrange plots in 1 row
common.legend = TRUE, # Use a common legend
legend = "bottom" # Place legend at the bottom
)
3.3 Nested Data
Nested data are as tricky as care trajectories. You have to handle the data properly and be careful in the level of detail that you want to display.
3.3.1 Treemap
Treemaps get the job done. They’re not quite as aesthetic as some other (curvy) options, but I feel like the fact they are squared makes it easier to read the data and “see” which area is bigger.
# Create a data frame containing cause of death, details (subcategory), and occurrences
dataTreemap <- data.frame(
cause_of_death = c(
"Cancer",
rep("Non cancerous diseases", 6),
rep("Road accidents", 2),
rep("Other or unknown causes", 4),
"Suicide"
),
details = c(
NA, # Subcategories for "Cancer" are not specified
"Cardiac", # Subcategories for "Non cancerous diseases"
"Respiratory",
"Mental disorders",
"Digestive",
"Endocrinian",
"Other",
"Car", # Subcategories for "Road accidents"
"Other",
"Weapons", # Subcategories for "Other or unknown causes"
"Poorly defined",
"Undefined",
"Other",
NA # Subcategories for "Suicide" are not specified
),
n = c(62, 19, 2, 1, 3, 2, 8, 7, 26, 5, 14, 8, 8, 52) # Occurrences for each category/subcategory
)
# Add the total count (n) for each cause_of_death to its label
dataTreemap <- as.data.frame(t(apply(dataTreemap, 1, function(x) {
x[1] <- paste0(
x[1], " (n=", # Append the total count to the cause_of_death label
sum(dataTreemap$n[dataTreemap$cause_of_death == x[1]]),
")"
)
x
})))
dataTreemap$n <- as.numeric(dataTreemap$n) # Convert the n column back to numeric after transformation
# Generate the treemap
treemap(
dataTreemap,
index = c("cause_of_death", "details"), # Define hierarchical levels for the treemap
vSize = "n", # Size of rectangles corresponds to the n column
type = "index", # Color rectangles based on the index
algorithm = "pivotSize", # Use pivot size layout algorithm for better space optimization
title = "", # No title for the treemap
align.labels = list(c("left", "top"), c("left", "bottom")), # Align labels in blocks
palette = "Pastel1", # Use pastel colors for the treemap
fontsize.title = 22, # Title font size
fontcolor.labels = "black", # Set label font color to black
bg.labels = 0, # Transparent background for labels
aspRatio = 1 # Set aspect ratio to make the blocks square
)